Ok, since we ran into so many problems last time around, why don't we
simply be less ambitious. Perhaps trying to solve so many things all
at once in such a delicate area was a huge mistake. :-)
To that end, let's start with just a LEA64 macro and convert
mpn/sparc64 over to it.
It uses rdpc, therefore it doesn't have to be mindful of PIC thunks
nor having to cater to leaf functions in a special way.
It also doesn't try to use gotdata relocations.
I have my fingers crossed, hoping that this approach has much less
fallout. But, we'll see. :-)
2013-04-15 David S. Miller <da...@davemloft.net>
* mpn/sparc32/sparc-defs.m4 (LEA64): New macro.
* mpn/sparc64/gcd_1.asm: Use it.
* mpn/sparc64/ultrasparct3/dive_1.asm: Likewise.
* mpn/sparc64/ultrasparct3/invert_limb.asm: Likewise.
* mpn/sparc64/ultrasparct3/mode1o.asm: Likewise.
diff -r 5ea92466ab67 mpn/sparc32/sparc-defs.m4
--- a/mpn/sparc32/sparc-defs.m4 Mon Apr 15 23:00:14 2013 +0200
+++ b/mpn/sparc32/sparc-defs.m4 Mon Apr 15 17:42:20 2013 -0700
@@ -46,4 +46,23 @@
define(`lzcnt', ``lzd' $1, $2')
')
+dnl Usage: LEA64(symbol,reg,pic_reg)
+dnl
+dnl Use whatever 64-bit code sequence is appropriate to load "symbol" into
+dnl register "reg", potentially using register "pic_reg" to perform the
+dnl calculations.
+
+define(LEA64,
+m4_assert_numargs(3)
+m4_assert_defined(`HAVE_GOTDATA')
+`ifdef(`PIC',`
+ rd %pc, %`$2'
+ sethi %hi(_GLOBAL_OFFSET_TABLE_+4), %`$3'
+ add %`$3', %lo(_GLOBAL_OFFSET_TABLE_+8), %`$3'
+ add %`$2', %`$3', %`$3'
+ sethi %hi(`$1'), %`$2'
+ or %`$2', %lo(`$1'), %`$2'
+ ldx [%`$3' + %`$2'], %`$2'',`
+ setx `$1', %`$3', %`$2'')')
+
divert
diff -r 5ea92466ab67 mpn/sparc64/gcd_1.asm
--- a/mpn/sparc64/gcd_1.asm Mon Apr 15 23:00:14 2013 +0200
+++ b/mpn/sparc64/gcd_1.asm Mon Apr 15 17:42:20 2013 -0700
@@ -94,22 +95,7 @@
L(noreduce):
-ifdef(`PIC',`
- rd %pc, %g3
- sethi %hi(_GLOBAL_OFFSET_TABLE_+4), %g4
- add %g4, %lo(_GLOBAL_OFFSET_TABLE_+8), %g4
- add %g3, %g4, %g4
- sethi %hi(ctz_table), %g1
- or %g1, %lo(ctz_table), %g1
- ldx [%g4+%g1], %i5
-',`
- sethi %hh(ctz_table), %l7
- or %l7, %hm(ctz_table), %l7
- sllx %l7, 32, %l7
- sethi %lm(ctz_table), %g1
- add %l7, %g1, %l7
- or %l7, %lo(ctz_table), %i5
-')
+ LEA64(ctz_table, i5, g4)
cmp %o0, 0
bnz %xcc, L(mid)
diff -r 5ea92466ab67 mpn/sparc64/ultrasparct3/dive_1.asm
--- a/mpn/sparc64/ultrasparct3/dive_1.asm Mon Apr 15 23:00:14 2013 +0200
+++ b/mpn/sparc64/ultrasparct3/dive_1.asm Mon Apr 15 17:42:20 2013 -0700
@@ -54,22 +54,7 @@
srlx d, 1, %g1
and %g1, 127, %g1
-ifdef(`PIC',`
- rd %pc, %g3
- sethi %hi(_GLOBAL_OFFSET_TABLE_+4), %g4
- add %g4, %lo(_GLOBAL_OFFSET_TABLE_+8), %g4
- add %g3, %g4, %g4
- sethi %hi(binvert_limb_table), %g2
- or %g2, %lo(binvert_limb_table), %g2
- ldx [%g4+%g2], %g2
-',`
- sethi %hh(binvert_limb_table), %g3
- or %g3, %hm(binvert_limb_table), %g3
- sllx %g3, 32, %g3
- sethi %lm(binvert_limb_table), %g2
- add %g3, %g2, %g3
- or %g3, %lo(binvert_limb_table), %g2
-')
+ LEA64(binvert_limb_table, g2, g4)
ldub [%g2+%g1], %g1
add %g1, %g1, %g2
mulx %g1, %g1, %g1
diff -r 5ea92466ab67 mpn/sparc64/ultrasparct3/invert_limb.asm
--- a/mpn/sparc64/ultrasparct3/invert_limb.asm Mon Apr 15 23:00:14 2013 +0200
+++ b/mpn/sparc64/ultrasparct3/invert_limb.asm Mon Apr 15 17:42:20 2013 -0700
@@ -35,25 +35,10 @@
srlx d, 55, %g1
add %g1, %g1, %g1
-ifdef(`PIC',`
- rd %pc, %g3
- sethi %hi(_GLOBAL_OFFSET_TABLE_+4), %g4
- add %g4, %lo(_GLOBAL_OFFSET_TABLE_+8), %g4
- add %g3, %g4, %g4
- sethi %hi(approx_tab), %g2
- or %g2, %lo(approx_tab), %g2
- ldx [%g4+%g2], %g2
- add %g1, -512, %g1
-',`
- sethi %hh(approx_tab-512), %g3
- or %g3, %hm(approx_tab-512), %g3
- sllx %g3, 32, %g3
- sethi %lm(approx_tab-512), %g2
- add %g3, %g2, %g3
- or %g3, %lo(approx_tab-512), %g2
-')
+ LEA64(approx_tab, g2, g3)
+ sub %g2, 512, %g2
+ srlx d, 24, %g4
lduh [%g2+%g1], %g3
- srlx d, 24, %g4
add %g4, 1, %g4
sllx %g3, 11, %g2
add %g2, -1, %g2
diff -r 5ea92466ab67 mpn/sparc64/ultrasparct3/mode1o.asm
--- a/mpn/sparc64/ultrasparct3/mode1o.asm Mon Apr 15 23:00:14 2013 +0200
+++ b/mpn/sparc64/ultrasparct3/mode1o.asm Mon Apr 15 17:42:20 2013 -0700
@@ -41,22 +41,7 @@
srlx d, 1, %g1
and %g1, 127, %g1
-ifdef(`PIC',`
- rd %pc, %g3
- sethi %hi(_GLOBAL_OFFSET_TABLE_+4), %g4
- add %g4, %lo(_GLOBAL_OFFSET_TABLE_+8), %g4
- add %g3, %g4, %g4
- sethi %hi(binvert_limb_table), %g2
- or %g2, %lo(binvert_limb_table), %g2
- ldx [%g4+%g2], %g2
-',`
- sethi %hh(binvert_limb_table), %g3
- or %g3, %hm(binvert_limb_table), %g3
- sllx %g3, 32, %g3
- sethi %lm(binvert_limb_table), %g2
- add %g3, %g2, %g3
- or %g3, %lo(binvert_limb_table), %g2
-')
+ LEA64(binvert_limb_table, g2, g4)
ldub [%g2+%g1], %g1
add %g1, %g1, %g2
mulx %g1, %g1, %g1
_______________________________________________
gmp-devel mailing list
gmp-devel@gmplib.org
http://gmplib.org/mailman/listinfo/gmp-devel
